In this thesis we consider the application of Sequential Monte Carlo (SMC) methods to continuous-time asset pricing models. The first chapter of the thesis gives a self-contained overview on SMC methods. In particular, starting from basic Monte Carlo techniques we move to recent state of the art SMC algorithms. In the second chapter we review existing methods for the exact simulation of Hawkes processes. From our analysis we infer that the simulation scheme of Dassios and Zaho (2013) outperforms the other algorithms, including the most popular thinning method proposed by Ogata (1980). This chapter serves also as introduction to self-exciting jump processes, which are the subject of Chapter 3. Hence, in the third chapter we propose a new self-exciting jump diffusion model in order to describe oil price dynamics. We estimate the model by applying a state of the art SMC sampler on both spot and futures data. From the estimation results we find evidence of self-excitation in the oil market, which leads to an improved fit and a better out of sample futures forecasting performance with respect to jump-diffusion models with constant intensity. Furthermore, we compute and discuss two optimal hedging strategies based on futures trading. The optimality of the first hedging strategy proposed is based on the variance minimization, while the second strategy takes into account also the third-order moment contribution in considering the investors attitudes. A comparison between the two strategies in terms of hedging effectiveness is provided. Finally, in the fourth chapter we consider the estimation of continuous-time Wishart stochastic volatility models by observing portfolios of weighted options as in Orlowski (2019). In this framework we don't know the likelihood in closed-form; then we aim to estimate it using SMC techniques. To this end, we marginalize latent states and perform marginal likelihood estimation by adapting the recently proposed controlled SMC algorithm (Heng et. Al. 2019). From the numerical experiments we show that the proposed methodology gives much better results with respect to standard filtering techniques. Therefore, the great stability of our SMC method opens the door for effective joint estimation of latent states and unknown parameters in a Bayesian fashion. This last step amounts to design an SMC sampler based on a pseudo-marginal argument and is currently under preparation.

In questa tesi si considera l’applicazione di metodi Monte Carlo sequenziali per modelli di asset pricing di tipo dinamico. Il primo capitolo della tesi presenta una panoramica generale sui metodi Monte Carlo sequenziali. Nello specifico, partendo da metodi Monte Carlo standard si giunge fino allo stato dell’arte per quanto riguarda i metodi Monte Carlo sequenziali. Il secondo capitolo costituisce una review della letteratura sui metodi di simulazione esatta per processi di Hawkes. Dall’analisi svolta si evince che lo schema proposto da Dassios e Zaho (2013) performa meglio degli altri algoritmi, incluso il più noto metodo “thinning” proposto da Ogata (1981). Questo capitolo serve inoltre come introduzione ai processi di salto di tipo auto eccitante, che saranno oggetto di studio del Capitolo 3. Nel terzo capitolo, quindi, viene proposto un nuovo modello diffusivo con salti auto eccitati per descrivere la dinamica del prezzo del petrolio. Il modello viene stimato implementando una recente metodologia di tipo Monte Carlo sequenziale utilizzando dati spot e futures. Dalla stima viene confermata la presenza di salti auto eccitati nel mercato del petrolio; questo conduce ad un migliore adattamento del modello ai dati e a migliori performance in termini di previsione dei futures rispetto ad un modello con intensità costante. Inoltre, vengono calcolate e discusse due strategie di copertura ottimali basate sul trading di contratti futures. La prima strategia è basata sulla minimizzazione della varianza, mentre la seconda tiene in considerazione anche la skewness. Viene infine proposto un confronto tra le due strategie in termini di efficacia della copertura Nel quarto capitolo si considera la stima di modelli a volatilità stocastica a tempo continuo basati su processi di Wishart, osservando portafogli di opzioni come in Orlowski (2019). In questo contesto la funzione di verosimiglianza non è nota esplicitamente, quindi verrà stimata ricorrendo a metodi Monte Carlo sequenziali. A questo proposito, i processi latenti vengono marginalizzati e la stima della verosimiglianza viene effettuata adattando metodi Monte Carlo sequenziali “controllati”, recentemente proposti da Heng et. Al. (2019). Dai risultati numerici si mostra come la metodologia proposta dia risultati decisamente migliori rispetto a metodi standard. Pertanto, l’elevata stabilità della metodologia proposta permetterà di costruire algoritmi per la stima congiunta di processi latenti e parametri utilizzando un approccio Bayesiano. Quest’ultimo step si traduce nel costruire un così detto SMC sampler, il quale è attualmente in fase di studio.

(2020). Application of Sequential Monte Carlo Methods to Dynamic Asset Pricing Models. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2020).

Application of Sequential Monte Carlo Methods to Dynamic Asset Pricing Models

GONZATO, LUCA
2020

Abstract

In this thesis we consider the application of Sequential Monte Carlo (SMC) methods to continuous-time asset pricing models. The first chapter of the thesis gives a self-contained overview on SMC methods. In particular, starting from basic Monte Carlo techniques we move to recent state of the art SMC algorithms. In the second chapter we review existing methods for the exact simulation of Hawkes processes. From our analysis we infer that the simulation scheme of Dassios and Zaho (2013) outperforms the other algorithms, including the most popular thinning method proposed by Ogata (1980). This chapter serves also as introduction to self-exciting jump processes, which are the subject of Chapter 3. Hence, in the third chapter we propose a new self-exciting jump diffusion model in order to describe oil price dynamics. We estimate the model by applying a state of the art SMC sampler on both spot and futures data. From the estimation results we find evidence of self-excitation in the oil market, which leads to an improved fit and a better out of sample futures forecasting performance with respect to jump-diffusion models with constant intensity. Furthermore, we compute and discuss two optimal hedging strategies based on futures trading. The optimality of the first hedging strategy proposed is based on the variance minimization, while the second strategy takes into account also the third-order moment contribution in considering the investors attitudes. A comparison between the two strategies in terms of hedging effectiveness is provided. Finally, in the fourth chapter we consider the estimation of continuous-time Wishart stochastic volatility models by observing portfolios of weighted options as in Orlowski (2019). In this framework we don't know the likelihood in closed-form; then we aim to estimate it using SMC techniques. To this end, we marginalize latent states and perform marginal likelihood estimation by adapting the recently proposed controlled SMC algorithm (Heng et. Al. 2019). From the numerical experiments we show that the proposed methodology gives much better results with respect to standard filtering techniques. Therefore, the great stability of our SMC method opens the door for effective joint estimation of latent states and unknown parameters in a Bayesian fashion. This last step amounts to design an SMC sampler based on a pseudo-marginal argument and is currently under preparation.
STEFANI, SILVANA
SGARRA, CARLO
SMC; Asset Pricing; Copertura ottimale; Pricing di opzioni; Econometria
SMC Methods; Asset Pricing; Optimal Hedging; Option Pricing; Econometria
SECS-S/06 - METODI MATEMATICI DELL'ECONOMIA E DELLE SCIENZE ATTUARIALI E FINANZIARIE
English
19-nov-2020
STATISTICA E FINANZA MATEMATICA
32
2019/2020
open
(2020). Application of Sequential Monte Carlo Methods to Dynamic Asset Pricing Models. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2020).
File in questo prodotto:
File Dimensione Formato  
phd_unimib_814542.pdf

accesso aperto

Descrizione: PhD thesis
Tipologia di allegato: Doctoral thesis
Dimensione 2.86 MB
Formato Adobe PDF
2.86 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/295144
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact